M
ASTER'
S DEGREE PROGRAMMESustainable Energy Systems
Software Models for Operating a Smart Grid Demonstrator with Respect to Load
Optimization
SUBMITTED AS A
M
ASTER THESISto obtain the academic degree of Master of Science in Engineering (MSc)
by
Khalil Dawi Musa Hamad 2016-12-07
Thesis supervisor FH-Prof. Dr. Peter Zeller
II
S WORN D ECLARATION
I hereby declare that I prepared this work independently and without help from third parties, that I did not use sources other than the ones referenced and that I have in- dicated passages taken from those sources.
This thesis was not previously submitted in identical or similar form to any other examination board, nor was it published.
...
Khalil Dawi Musa Hamad Wels, December 2016
A CKNOWLEDGMENT
First I would like to express my gratitude to my supervisor Dr. Peter Zeller for giving me the chance to work on this topic as a part of his research team. This work would have not been possible if not for his continuous support, patience, and enthusiasm. I also want to thank all the professors who taught me during the past two years of my study in FH OÖ Wels campus, and all the staff members of the FH especially Ms. Daniela Hochstöger the administrative assistance for her patience and tremendous help through- out my study period. I also want to extend my thanks to my colleagues Andres Moreno and Mohammadhossein Sharifi who were a great team to work with during the past year. I also want to give thanks to Eaton Corporation represented in Mr. Wolfgang Hau- er and Mr. Michael Bartonek for supporting us with necessary hardware to realize our work. Last but not least I would like to thank my friends and family especially my par- ents Asha Elkarib and Dawi Hamad for supporting me throughout all stages of my life.
IV
K URZFASSUNG
Mit der Einführung von Smart-Grid-Konzepten, könnten die Kohlendioxidemissionen deutlich reduziert werden, indem der Weg für die Integration erneuerbarer Energiequel- len im Stromnetz frei gemacht wird. Das Smart Grid ist ein Upgrade im bestehenden Stromnetz. Beim traditionellen Stromnetz fließt die Energie nur vom Ort der Erzeugung zum Abnehmer. Das Upgrade besteht darin, eine Zwei-Wege-Kommunikation einzu- binden, um eine beidseitige Energiespeisung zu ermöglichen. Mit einer derartigen Kommunikation kann neben der Einspeisung von Energie auf Verbraucherseite ein be- stehendes Netz höher ausgelastet werden, indem man über die zusätzlich installierte Kommunikation Spitzenlasten verringert (Load Shift).
Das Ziel dieser Masterarbeit ist es, das Potenzial von Lastautomatisierung speziell in Haushalten zu untersuchen, sowie Softwarekomponenten zur Steuerung eines Smart Grid Demonstrators zu entwickeln. Der Demonstrator soll den Kommunikationsweg und die Steuerung im Haushalt selbst umfassen. Die in dieser Arbeit untersuchten Las- ten sind eine Wärmepumpe, ein Elektroauto, eine Waschmaschine, ein Wäschetrockner, eine Geschirrspülmaschine, eine Fotovoltaikanlage sowie weitere Grundlasten (die mit- tels eines Leistungsprofiles berücksichtigt werden). Die entwickelte Software umfasst Kontrollalgorithmen sowie die Emulation von realistischem Lastverhalten. Die Modelle sind basierend auf LabVIEW entwickelt. Im Demonstrator wird Hardware-in-the-loop Simulationstechnik angewandt. Oben genannte Haushaltslasten sind basierend auf Be- triebsgrenzen und Statistiken von Nutzerverhalten modelliert. Die Lasten werden auf- grund des solaren Angebotes, sowie auf Netzvorgaben (die mittels drei Zuständen kommuniziert werden) gesteuert.
In der Arbeit kann gezeigt werden, dass mit Hilfe eines intelligenten eine Erzeugungs- Lastanpassung erfolgreich durchgeführt werden kann. Es zeigte sich, dass 40% der Spitzenlast – in den betrachtenden Haushalten –durch eigentlich kontrollierbare Lasten verursacht werden, wodurch ein großes Potential zur Lastverschiebung besteht. Es konnte gezeigt werden, dass die entsprechende Hardware (Kommunikation / Steuerung /
intelligente Schalter), basierend auf den Netzvorgaben die Spitzenlast erfolgreich sen- ken kann (um 40 % bis 95 %).
Die Zukunft solcher intelligenten Lösungen hängt von der Entwicklung der Smart-Grid- Standards auf Haushaltsebene sowie von der kundenseitigen Bereitschaft und Nachfrage ab.
VI
A BSTRACT
With the adoption of smart grid concepts, carbon dioxide emissions could be reduced significantly by paving the way toward the integration of more renewable generation in the electricity grid. The smart grid is an upgrade of the existing electricity grid that tra- ditionally delivers electricity from generation plants to final customers mostly in one- way power flow. The upgrade is to include two-way communication technologies to accommodate power flows from distributed generation. With such a communication, the existing network can be heavily utilized by reducing of consumers peaks by means of load shifting.
The aim of this thesis is to investigate the potential of load automation in households and to develop software models for operating a smart grid demonstrator. The demon- strator includes components used for home automation with the communication needed between those components. The loads covered in this thesis are a heat pump, an electric vehicle, a dishwasher, a tumble dryer, a washing machine, photovoltaic generation and other base loads. This is done by developing software load control algorithms that re- flect control potential with realistic load profiles. The models are developed in Lab- VIEW, and the demonstrator uses Hardware-in-the-loop simulation technique. The aforementioned loads are modelled based on operation constraints and user behavior statistics. The loads are controlled depending on the photovoltaic generation availability and grid information which are represented with three grid levels.
This work shows that with the help of intelligent switching components, loads can be shifted in households according to generation status. The thesis concluded that approx- imately 40% of peaks in the considered household are caused by controlled loads, which offers a good potential for load shifting. The corresponding components of the demonstrator (communication, control, and intelligent switches) can successfully reduce peaks caused by controlled loads by 40 % to 95 % depending on the grid requirements.
The future of such solutions relies on the development of smart grid standards for household levels and on the willingness of customers to participate in load shifting.
C ONTENTS
Kurzfassung ... IV Abstract ... VI
1 Introduction... 1
1.1 Background... 1
1.2 Electrical Grids Current Status ... 2
1.3 Smart Grids Today ... 4
2 Research Background ... 9
2.1 Potential in Households ... 9
2.2 Smart Households Research State of the Art ... 10
2.3 Research Gaps ... 18
2.3.1 Full System Demonstrator ... 18
2.3.2 Grid Information Model ... 18
2.3.3 Consumption Optimization Based on a Grid Information Model ... 18
2.4 Scientific Question ... 20
2.5 Scientific Approach ... 21
2.6 Delimitations ... 21
3 Software Models Description ... 22
3.1 Software Models Overview ... 23
3.2 Heat Pump Mathematical Modelling... 28
3.3 Heat Pump Constraint... 33
3.4 Washing Appliances Mathematical Model ... 35
3.5 Washing Appliances Constraints ... 36
3.6 Electric Vehicle Mathematical Model ... 38
3.6.1 Charging Location ... 38
3.6.2 Charging Moments ... 39
3.6.3 Charging Need ... 40
3.7 Electric Vehicle Constraints ... 42
3.8 Base Load Model ... 44
3.9 Photovoltaic Data ... 47
VIII
3.10 Load Control Algorithm ... 48
4 Results and Discussions ... 53
4.1 Long-term Results ... 53
4.2 Energy Optimization Short-term Results ... 61
4.2.1 Case Study One ... 61
4.2.2 Case Study Two ... 65
4.2.3 Case Study Three ... 69
5 Conclusion and Future Work ... 73
5.1 Conclusions ... 73
5.2 Future Work... 74
6 List of abbreviations ... 77
7 List of Figures ... 79
8 List of tables ... 81
9 Bibliography ... 82
10 Appendix ... 88
10.1 Load Models Statistical Data... 88
10.2 Photovoltaic Readings ... 93
1 I NTRODUCTION
1.1 BACKGROUND
Picture a future where all energy needs are realized in a sustainable way. This would mean that the dependency on fossil fuels would be reduced significantly. This would be a future where global warming is reduced, where economic growth is achieved with innovative and sustainable use of natural resources. To achieve such a future three main actions are to be acknowledged: reduction of energy consumption, improvement of en- ergy efficiency, and production of more energy from renewable sources. The reduction of energy consumption should have the priority. This would cause both energy produc- ers and consumers to actively participate in energy optimization. Smart grids would be in the frontline of this future.
To combat climate change the European Union (EU) energy polices reflects energy effi- ciency as a key component combined with promotion of renewable energies. Increasing renewable energy sources (RES) share in electrical grids is becoming an important topic in the development of new energy infrastructures (see Fig. 1). The European Commis- sion has set a climate and energy framework to cut at least 40% in greenhouse gas emis- sions (from 1990 levels), at least 27% share increase of RESs, and at least 27% im- provement in energy efficiency by 2030 [1]. The EU is also committing to reduce greenhouse gas emissions by 2050 to 80-95% below 1990 levels. To achieve such ambi- tion goals, which are based on uncertain assumptions, great structural and social efforts are required. Changes in oil prices, market changings and the difficulty to predict the successes of new technological advancements are the main causes for uncertainty. One of the goals is to increase electricity consumption share from RES to above 95%. To achieve such a goal “the (electrical) distribution grid needs to become smarter to deal with variable generation from many distributed sources […] (and) also increased de- mand response.” [2, p. 6]. Demand response (DR) is a supply/demand balancing mech- anism in which customer’s energy consumption responds to changes in supply condi- tions because of the less predictable supply of wind and solar generations, compared to fossil or nuclear generation [3].
According to the International Energy Agency (IEA) non-hydro renewable electricity generation in Europe increased by 14% to reach 640 TWh in between 2005-2015 higher than hydro electricity generation (564 TWh) for the first time [4]. In 2014 the total en- ergy produced in Austria was 65.4 TWh from which 5% was produced from wind and around 1% from solar photovoltaic (PV). Wind and solar PV energy produced increased from 1352 to 4631 GWh in between 2005 and 2014 [5].
Fig. 1: EU De-carbonization scenarios - range of fuel shares in primary energy con- sumption compared with 2005 outcome (in %) [2].
1.2 ELECTRICAL GRIDS CURRENT STATUS
Today’s electrical grids are built to accommodate unidirectional power flow from large generation plants through transmission and distribution systems to the end consumers.
The traditional grid system is operated with high degrees of reliability and operators are familiar with variability in load consumption. Nevertheless, operation is becoming more complicated with the penetration of renewables in the electrical grid. High penetration of renewables causes unpredictable generation, new power flow directions, voltage fluc- tuations, and stress on electrical grid protection devices [3].Currently different ap- proaches are used to address these issues such as grid reinforcement by adding new components, transformers on line tap changers (OLTC), and reactive power control
techniques. To maintain thermal and voltage constrains on distribution and transmission lines more measurement and monitoring devices have to be added to the grid and a two way communication path between prosumers and grid operators becomes essential [2, 3, 6].
Currently, the electrical grid operators maintain a precise balance between supply and demand within a certain geographical area known as a balancing region. The balancing regions together form a control area. In a control area, the generators are dispatched to balance the change in load, which is a supply-side management technique. For abnormal situations in a control area, some generation capacity is available to the grid operator within short time to meet the balancing requirement. This generation capacity is known as the operating reserve capacity of the grid, and is divided into primary, secondary and tertiary control reserves. Primary reserves operate automatically and are referred to as frequency response reserves. Secondary reserves operate also automatically after around 30 seconds, and use the spinning reserves of rotating generators in the grid. Tertiary control on the other hand is operated manually by the grid operator and is considered as the non-spinning reserve of the grid. The Austrian Power Grid (APG) for example has reserves of ±70 MW for primary, ±200 MW/+280MW for secondary, and +280 MW / - 150MW for tertiary controls. Changes in wind speeds or passing clouds would change the output of renewables, and other dispatchable generators or loads have to cover for that. Hence, the integration of renewables will add to the reserve capacity requirements of the grid. The electrical grid should have enough operating reserve capacity from in- terconnections, storage, backup supply, or fixable loads controlled by DR mechanisms.
DR requires a two ways communication, metering, and real-time monitoring i.e. a smart grid. [3, 7].
Electricity storage can be categorized into short and long-term storage. For large scale long-term seasonal storage, hydro power can be utilized to store electricity, but this is limited by natural resources available. Short-term measures such as batteries, pumped storage, compressed air and flywheels can also provide a flexible reserve capacity.
However, as a consequence out of the lack of feasible storage of electricity, it’s better to consume the electricity produced from renewables as it is being generated. DR is an
PV-generation is to some extend cyclic and correlates with daily load profiles, especial- ly to cover noon peaks. However, PV produces energy which is variable and less pre- dictable according to bad weather conditions where reduce PV-generation is observed.
Wind generation on the other hand is even more unpredictable and does not follow daily load profiles. An inverse correlation is sometimes observed in which wind generation on average is occurring during hours of limited demand at night. For both PV and wind generations, it is difficult to forecast short-term weather conditions and only mid-term weather forecast can be useful. Energy markets need to add measures to accommodate the variability of renewable production in the dispatching process [3, 8].
1.3 SMART GRIDS TODAY
The different concerns about large scale penetration of renewables at both local con- sumer and grid level can only be dealt with appropriately by measures of smart grids.
According to the European commission report on smart grids, they could be defined as
“an upgraded electricity network to which two-way digital communication between supplier and consumer, intelligent metering and monitoring systems have been added”.
The European smart grid Task Force defines smart grids as “electricity networks that can efficiently integrate the behavior and actions of all users connected to it […] in or- der to ensure an economically efficient, sustainable power system with low losses and high quality and security of supply and safety” [8, p. 2]. Smart meters play a considera- ble role in smart grids. EU plans by 2020 to replace at least 80% of electricity meters to smart meters. A recent report from the same reference expects that 72% of meters would be replaced by smart meters [9] . However a smart grid is much more than that.
Fig. 2: Power and data flows in smart grids, based on [38].
In (Fig. 2), a diagram shows power and data flows for a smart grid compared to a tradi- tional one. New renewable plants are added in all grid levels, mostly in medium voltage (MV) and low voltage (LV) levels. Smart grids require having smart substations espe- cially as MV/LV stations. Those substations will get measurements data from added new sensors, smart meters, and customer energy management systems (CEMS) and would provide a gateway to the grid operator. Also market information has to be availa- ble to final customers. Customers from private or industry sectors would require to have switching actuators with energy measurements and protection (SAEMP) either with integrated solutions like in [10], or as additional measurement/ actuators to the already existing protection devices. [3, 11, 9].
Interoperability is a key factor in smart grids to improve the needed decision-making in a smart grid with more complexity. The addition of new hardware and software to en- hance the grid operation brings much capability requirements. Smart grid interoperabil- ity means the secure and effective use of information shared between different compo- nents and users of the grid system. The smart grid Coordination Group (SG-CG) Refer- ence Architecture Working Group (SG-CG/RA) developed a framework for standardi- zation of smart grid application. A smart grids architecture model (SGAM) (see Fig.
3) was presented by this group which is a three-dimensional model for the different in- teroperability layers with the two dimensions of smart grid layers of domains and zones [12, 13].
Fig. 3: Smart grids architecture model framework [12].
The SGAM models (see Fig. 3) shows that in a smart grid system, different products from different manufacturers should work all together. Also the end customer will take more roles by participating in the energy market. The smart grid zones cover the hierar- chal levels of power system management, while the domains cover the complete electri- cal energy conversion chain [12].
Adopting the concept of smart grids offers many benefits. They are essential for inte- grating renewables to electrical grids while avoiding high investment on grid reinforce- ment. Smart grids can manage interconnections of all electrical grid users from consum- ers, energy suppliers, and market participants. They also provide bases for incentivized energy efficiency with the combination of price information [8]. The addition of state of the art information and communications technologies (ICT) to the current electrical grids adds to the system’s reliability, “Improved and more targeted management of the grid translate into a grid that is more secure and cheaper to operate”. Moreover, smart grids would provide a platform for future innovation in energy services while insuring data protection and cyber-security challenges [8, pp. 2-3].
Smart grids still face serious challenges. The main challenge in Europe is developing common European smart grid standards. Regulatory frameworks also represent con- cerns as there is still uncertainty on cost/benefit sharing between different grid users.
Hence, replicability of projects in different countries is still a barrier [8, 14], [14] also lists unwillingness of customers to participate in trails as a hindrance to smart grid pro- jects realization.
2 R ESEARCH B ACKGROUND
2.1 POTENTIAL IN HOUSEHOLDS
Buildings offer great potential for smart grid applications on the distribution level of the electrical grid. Commercial, public, and residential buildings consume about 27% of final electricity in Europe with a saving potential of 30% by 2020 compared to 2005 consumption [15]. Most of this electrical consumption is used for heating and lighting.
According to EU commission, “Energy management systems can greatly reduce the CO2 footprint of buildings”. EU is placing actions on industry to develop intelligent home control systems for household applications to enhance loads management and temperature control in residential buildings [15, p. 9]. According to European commis- sion on buildings “by improving the energy efficiency of buildings, we could reduce total EU energy consumption by 5% to 6% and lower CO2 emissions by about 5%” [16, p. 1]. In Austria, residential and public buildings consumed 29% and 21% of final elec- tricity in 2014 respectively [4].
Household loads can be categorized to controllable and non-controllable loads (see Fig.
4). Considering home appliances with regards to demand response functionality or smart management some loads can be controlled automatically like electric vehicles (EV), heat pump (HP), air conditioner (AC), tumble dryer (TD), dishwasher (DW), washing machine(WM), and to some extend fridges and freezers. Controllable loads are flexible in operation and have small impacts on consumer lifestyle. Other loads are crit- ical non-controllable loads such as lights, television, iron, and cooking appliances. The percentage of controlled loads was found to be approximately 50% of household’s elec- tricity consumption. Compared to the total country consumption there would be almost 8 TWh of controllable energy, which represents 14% of the total energy consumed in one year. With the expected addition of more electric vehicles this percentage will go higher. This represents a good potential for peak load shifting of households.
Fig. 4: Controllable load power consumption in Austrian households 2012 based on statistics from [17].
According to a research done on scenarios of demand side energy management concepts in Austria [11], communication between loads in a household and the PV generation system could help avoiding the strict requirements on PV generation. Flexible loads can be utilized to avoid load peaks without introducing additional storage units. The re- search [11] also found that micro grid scenarios for buildings are among the most prom- ising scenarios in Austria for the next ten years.
2.2 SMART HOUSEHOLDS RESEARCH STATE OF THE ART
The residential sector represents a good potential for energy savings that energy provid- ers and other stakeholders in the smart grid business can utilize. A smart household is considered typically to have various appliances, local PV generation with or without storage, an electric vehicle, and some sort of home automation system. Household ap- pliances can be categorized to controllable and non-controllable appliances based on the potential of smart management (see page 9).
Currently, academy and industry sectors are the main players in smart homes [13]. In the smart home scientific field many sub-categories exist (see Table 1) and the activities are always overlapping in those projects. One project can tackle demand response and temperature control but without addressing power quality issues and so on. According
to smart grid laboratory inventory by the EU commission energy management, integra- tion of renewables, and demand response activities are the most addressed ones in Eu- rope followed by temperature control [13]. Another survey by the EU commission on 2014 found that out of around 460 projects in the smart grid fields more than 145 had smart customers as their main application [18]. Those figures reflect the importance of the smart consumer/prosumer behavior for the future of smart grids as this field is still filled with uncertainties.
Table 1: Activities regarding Smart Household research [13].
Field Percentage of field in
surveyed researches
Temperature Control 72%
Lighting 28%
Movement Sensors 23%
Power Quality 39%
Smart Appliances 61%
Security 11%
Safety 23%
User Account and Billing 17%
Demand Response 77%
Energy Management Strategies / Cost-control 84%
Interoperability 56%
Integration of RES 84%
The focus in this thesis is on research done on single home schemes rather than single home and community level or smart city schemes. Smart home modelling can be divid- ed to four main topics (see Fig. 5):
• Optimization methods.
• Scheduling techniques.
• Price schemes.
• Load modelling.
With regard to optimization methods different researchers have used different tech- niques. In (Fig. 5) different considerations for optimizing methods are shown. First the problem, object function, and constraints have to be defined exactly and then the proper approach can be chosen. For example we could have a mixed-integer problem where the function is linear and is heuristically approached. Optimizing methods can be generally classified to exact and heuristic methods. In exact methods there is a guarantee to have the optimum solution where in heuristic methods no such guarantee is provided. Heuris- tic methods are used when the effort for solution grows exponentially with the problem, and these methods are problem specific and they exploit the properties of that problem [19].
In [20] large number of scenarios are generated by stochastic variables and then a heu- ristic optimization algorithm is used to choose the optimal scenario (PSO). A day ahead robust scheduling of four loads is proposed. The loads and generations considered are:
• Plug-in hybrid electrical vehicle.
• Space heater.
• Water heater storage.
• Pool pump.
• PV generation of 2 kWp.
The energy price scheme considered is a time of use rate scheme and optimization is based on the consumers benefit and not on electric energy. In [21] a similar PSO opti- mization method is used to determine energy from different generators base on load status. The algorithm is based on real-time system to perform peak load shaving. Loads considered here are the followings:
• House heating system.
• Water heater.
• Washing machine.
• Tumble dryer.
Appliances are represented by interfaces with certain inputs such as start/stop signals and outputs such as status and energy required. The appliance is then operated with on/off signals from the algorithm output. The authors recommend that statistical consid- erations about usage of appliances could enhance the system. In [22], also a PSO opti- mization method is used.
Fig. 5: Related research areas for smart home modelling [23, 24, 25].
A Genetic Algorithm (GA) is used in [26] for load scheduling to shift load in order to minimize peaks in consumption of a household. A SCADA system for a household is applied here and MATLAB is applied for the load control algorithm. The algorithm is based on a specific power consumption limit determined by status of micro generation forecast and/or energy tariff. Loads and generation types considered in this study are:
• PV system.
• Wind turbine.
• Fuel cell.
• Induction motors.
• Dishwasher.
• Washing machine.
• Refrigerator
• Heat accumulators.
A mixed integer non-linear programming approach is used as a comparison. Authors in [27] use a mixed integer nonlinear optimizing programming (MINLP) algorithm to re- duces energy consumption costs and maintain customers comfort. In [28] a mixed inte- ger linear programming (MILP) to schedule loads based on electricity cost. Customer preferences and technical constraints are used in scheduling the loads. Loads considered in this study are:
• Dishwasher.
• Washing machine
• Tumble dryer.
In [29] a bottom up approach for modelling of household appliances is presented and used to generate an aggregated load profile. The models are developed on appliance level considering physical models with operational constraints. The models are validat- ed using real electricity consumption data. A load profile for non-controllable loads is created based on the historical data. Controllable loads considered in this study are:
• Space cooling/ heating equipment.
• Water heater.
• Tumble dryer.
• Electric vehicle.
In [23] a real-time home energy management combined with forecasted price infor- mation is developed. The loads and generations considered here are:
• Water heater.
• Tumble dryer.
• Air conditioning unit.
• Electric vehicle.
• Other critical loads
• PV system with battery.
Each load in [23] is built with constraints of operation. For energy prices a grid power limit rate scheme is applied and an agreement with energy provider is assumed where the household is informed about DR periods. The energy management algorithm aims to shift the energy consumption from high price time slots based on forecasted infor- mation. The battery is charged in low price periods and discharged in high price peri- ods. A Physical test platform is built in a lab environment with PV simulator system with a battery, three AC dynamic loads, and measurement and control units. The energy management system is implanted in a PC and is split into two parts, a monitoring and control part based on C#- and a decision making part based on MATLAB. The authors state that demand response to shift loads on for households might lead to new peak load periods and suggest further studies on flexibility of demand response.
In [30] a linear sequential optimization (LSO) enhanced algorithm is implemented to minimize the customer payments or maximize customer preferences. The focus here is on unit commitment issue and an electric water heater is used to describe the problem.
A physical model of the heater is developed with a statistical model of random water consumption. Day ahead scheduling is coupled with real-time adjustments to overcome the problem of uncertainties in real-time price forecasting. These studies and more are summarized in (Table 2).
Table 2: Research areas summary
Research area Publications
Optimization Method PSO [20, 21, 22], GA [26, 31], MINLP [26, 27], MILP [28], LSO [30], Convex programming [32]
Objective focus and limitations Customer benefit not grid [20],Peak load shaving [21], Power Limits [26],load profile modelling [29], Cost reduction [23, 27, 28]
Simulation Platforms1 MATLAB [20, 23, 26], Simulink state-flow toolbox [21], SCADA [26], C# [23]
Number of loads 4 loads/ battery storage [23],4 loads [20, 21, 29], 3 generators/induction motors/variable loads [26], 3loads [28], 1 load [30]
Some smart grid projects are implemented in the few past years in Austria. In ZUQDE project [33], central voltage and reactive power control techniques are implemented in Lungau region in Salzburg. A power controller is used to control transformers, loads, and renewable generators. Another project DG DemoNet [34] was implemented in Köstendorf to test control concepts for active LV network. A model community in Salzburg-Flachgau is chosen with 192kWp installed rooftop PV power in addition to EV and electrical heating systems in houses. The goal is to maximize utilization of lo- cally produced PV energy. A central building management system receives information from the LV sub-station and controls the house loads with the help of weather forecast data. In AMIS project [35] a concept for exchanging data between grid operator and smart meters is researched to give recommendation on a new business model and speci- fications of a new generation of smart meters, and over 90,000 meters were installed in Upper Austria. Other projects are still in implementation phase like hybrid-VPP4DSO, Smart Village Regau, Smart City Project Graz Mitte, and INIGRID which deal with different aspects in smart grid fields [36, 37].
1 Some publications use more than one platform for different purposes
The main components used in the application of DR mechanisms in smart households can be summarized in (Fig. 6). Some information is needed from the grid side in addi- tion to market information. In the household premise, a customer energy management system (CEMS) is required and some components are needed to control the loads such as SAEMPs described in (chapter 1 under Smart Grids Today).
Fig. 6: Typical smart household components, based on [38].
2.3 RESEARCH GAPS
The previous research shows the importance of developing accurate models in order to estimate load profiles of households. Smart energy management systems are being de- veloped by stakeholders in industry and research, but there is still a need for developing advanced control methods to integrate renewables in real-time to the electrical grid [39].
Research on load control algorithms that integrate both grid information and customer preferences is becoming essential.
2.3.1 FULL SYSTEM DEMONSTRATOR
According to smart grid Austria technology roadmap, there is a lack of research in ener- gy management systems [36]. Also long term impacts of smart household energy opti- mization are missing, especially ones that consider buildings boundaries. The report comments also on the lack of smart grid demonstrators in the academic level in Austria.
Only little work was done to build physical smart grid demonstrators that can be used a testing platform for real smart grid components for household levels. Such demonstrator will follow EU and Austrian polices to raise awareness of smart grids in both academic and public levels.
2.3.2 GRID INFORMATION MODEL
A traffic light model reflecting the status of the electrical grid was introduced first in Germany by E-energy program [40]. Cooperation between DACH-co and Austria helped to identify what each level means. There are still some uncertainties in defining exactly what each level means especially the Yellow level [36]. INTEGRA project is considered one of the first projects researching the requirements to define functionalities [41]. However, there is a lack of research on implementing this model in energy man- agement systems. It is important to develop load control algorithms to reflect what the grid needs at any moment.
2.3.3 CONSUMPTION OPTIMIZATION BASED ON A GRID INFORMATION MODEL
Some of the work in household energy management areas has a focus on mathematical modelling problems and the solutions are chosen based on specific simulation cases.
Little work was done to reflect a realistic scenario problem and a tradeoff of price, grid power limits, and customer preferences [24]. Most research focuses on scheduling based on forecast information and more real-time approaches have to be researched and pre- sented. There is a need to have a smart grid environment where loads are presented in a modular form and real hardware can be integrated easily with the energy management system. Also only little work is done on the requirements needed from household loads to be smart grid ready. This would give good recommendation to the manufacturers of loads and appliances to adapt the required interfaces to their products.
2.4 SCIENTIFIC QUESTION
The purpose of this thesis is to develop software models for the realization of a smart grid demonstrator for household’s level. The demonstrator provides a physical envi- ronment used for both testing real smart grid components and demonstration purposes.
The objectives are:
• To develop load models to reflect a realistic load profile for a household, such as:
• Heat Pump.
• Dishwasher.
• Tumble Dryer.
• Washing Machine.
• Electric vehicle.
• To develop load control algorithms to shift the load flow in the household based on the traffic light grid information model while considering local PV generation and customer preferences.
• To use the models in the smart grid demonstrator to perform an accelerated long-term simulations covering the period of one year and to investigate the im- pacts of real time demand response. The results would reflect households load profiles, load shifting possibilities and potential of energy optimization.
2.5 SCIENTIFIC APPROACH
A Hardware-in-the-loop (HIL) programming technique is implemented in LabVIEW [42] to build a real-time model. Accelerated simulation algorithms are developed to investigate a one year time period. Realization of software models is approached as fol- low:
• Thermal household model to estimate temperature changes inside the household based on outside conditions.
• Heat Pump (HP) model. Load profile is based on heat pump size and the thermal household model.
• Electric vehicle (EV), washing machine (WM), tumble dryer (TD), and dish- washer (DW) models. Load profile is based on behavior statistics in Austria.
• Base Load (BL). Load profile based only on critical non-controllable loads in a typical household such as lights, oven, cooker, and TVs.
• Photovoltaic (PV). Generation profile is driven from historical recordings of generated power. The readings are taken from the University of Applied scienc- es in Wels PV and meteorology data recording system.
• Load control algorithm: The problem is defined as an integer liner programing problem (ILP). A heuristic optimization method is used to approach it. Decisions to dispatch loads and generators are made every 15 minutes. For more infor- mation about related modelling techniques (see Fig. 5).
LabVIEW is chosen as the modelling software due to friendly user interfaces for data visualization and ease of communication of measurements through different bus sys- tems or webservers.
2.6 DELIMITATIONS
Communication concepts and hardware consideration for building the smart grid de- monstrator are not considered in this thesis and are covered by [43] [44] respectively.
3 S OFTWARE M ODELS D ESCRIPTION
This section will present an overview of the components used in the smart grid demon- strator. After, a detailed description of each software model is presented.
The layout of the developed smart grid demonstrator is shown in (Fig. 7). The figure shows the different hardware and software components used, In addition to the power and data flows. The focus in this thesis will be on the software models developed in PC 1. For more information on the communication and hardware aspects please refer to [43, 44] respectively.
The components of the demonstrator are:
• PC 1 containing CEMS and load software models. The load models include HP, DW, TD, WM, EV, PV and BL models.
• PC 2 containing smart meter and grid information software models. The grid in- formation is based on the traffic light model.
• SAEMP distribution board containing actuator and measurement unit hardware.
• SAEMP RF-gateway hardware to communicate wirelessly with the actuators and measurement units.
• Dynamic electronic loads. Used to emulate load consumption of actual house- hold loads.
Fig. 7: Smart grid demonstrator overview [44].
The flow of information in (Fig. 7) is as follow:
• The CEMS model (in PC 1) fetches the grid information from the grid infor- mation model (in PC 2).
• The CEMS model (in PC 1) requests the actuator statuses and the energy meas- urements from SAEMP gateway for the three connected dynamic loads.
• The CEMS model (in PC 1) fetches the photovoltaic and base load simulated in- formation from the models developed.
• The CEMS model (in PC 1) fetches the customer preferences from the load models.
• The CEMS model (in PC 1) uses the above information to generate the new sta- tus of the loads and PV (on, off) based on the control algorithm.
• The CEMS model (in PC 1) sends the switching results to the SAEMP gateway.
• The load models running in PC1 send the power consumption to the dynamic loads.
• Each cycle represents a15 minutes interval. The simulation is made in an accel- erated manner where each cycle takes on average about 35 seconds; hence 1 year could be simulated in 14 days [43].
3.1 SOFTWARE MODELS OVERVIEW
In this section, the software models located in PC 1 (see Fig. 7) are presented. The gen- eral overview of those models is shown in (Fig. 8), which shows the different infor- mation passed between the components in the demonstrator. (Fig. 8) only shows the flow of information and not the energy flow, and the components indicated are:
• Customer preferences software model.
• Load software model.
• Load constraint software model.
• Load control algorithm software model.
• Grid information software model.
• SAEMP hardware.
Fig. 8: Software models and information overview, development of dotted components are not considered in this thesis.
The CEMS is realized in two parts, the load constraints and the load control algorithm models. The load constraints represent the operational restrictions of each load and are different depending on the load. The data exchanged between the components in (Fig.
8) are:
• Time Remaining (tLrem) is the remaining time for the load to finish the task.
• Run Status (RSLt) is the status each load can have at each time step (see Table 4).
There are three options the load can have: must be OFF (RSLt = 0), must be ON (RSLt = 1), or CAN run (RSLt = 2). This is decided by the CEMS based on cus- tomer preferences and loads operation constraints.
• Grid information (GI) is the information from the grid operator. There are two types of GI: grid power limits which are the maximum (Pmaxt ) and minimum (Pmint ) active power limits at a certain time, and grid status level (GSt) which is represented with a three states traffic light model: green (GSt = 0) means no re- strictions from the grid, yellow (GSt = 1) means power limits should be fol-
lowed and if exceeded additional charges for breaking the limits, and red (GSt= 2) where the power limit must be strictly followed.
• Operation status (OPLt) is the operation status of the SAEMP connected to the load. Will be set be the CEMS; on (OPLt = 1) or off (OPLt = 0).
The load models get synchronization information (time and date) to simulate the new behaviour. The SAEMP hardware gets control signals from the CEMS and sends meas- urements to the CEMS. The dotted boxes represent the components which are not in- cluded in this thesis (details in [44], [43] ).
Below are important definitions for the realization of the software models.
• Customer Energy Management System (CEMS) is the main system that per- forms the energy optimization is made by controlling the operation of loads based on grid information and customer preferences.
• Switching actuator energy measurement and protection device (SAEMP). Each load is connected through an SAEMP that provides controlling possibility.
• The load constraint model is a realization of how the CEMS will keep the opera- tion constraints of each load. The constraints are different for different loads, the heat pump (HP) has temperature constraints (minimum and maximum tempera- tures to keep) while the electric vehicle (EV), WAs have task time constraints (charge the EV or finish a washing cycle within a time limit). It is also a way to describe different loads in 2 comparable parameters Run Status (𝑅𝑅𝐿𝑡) and re- maining time to finish the task (𝑡𝐿𝑟𝑟𝑟). The algorithm then prioritizes different loads based on those comparable parameters and takes the final decision. The fi- nal decision is to switch on or off the CLs.
• Task time (𝑡𝐿𝑡𝑡𝑡𝑡) is the required working time for the load, L stands for the load.
• Commitment window (𝐶𝐶𝐿) is the maximum time in which the customer prefers the task to be finished when activating DR mode. For example, the customer can specify a 7 hours CW to complete a laundry round that has a task time of two
hours. The CEMS will choose the optimum 2 hours to run the WM within the 7 hour CW that the customer gave.
• Start time (𝑡𝐿𝑡𝑡𝑡𝑟𝑡) is the time that the loads is switched on, but not necessarily when the load starts operating. When DR is activated by the customer it is the beginning of the (𝐶𝐶𝐿).
• End time (𝑡𝐿𝑟𝑒𝑒) is the time that the loads is switched off, not necessarily when the load stops operating. When DR is activated by the customer it is the ending of the (𝐶𝐶𝐿).
• Demand response commitment status (𝐷𝑅𝐶𝑅𝐿) is the status of DR for the load.
For each load the customer is given an option to run the load immediately (𝐷𝑅𝐶𝑅𝐿=0) or to activate DR commitment (𝐷𝑅𝐶𝑅𝐿=1) with a certain CW.
• The dishwasher (DW), Tumble dryer (TD), and the washing machine (WM) have similar characteristics and will be referred to as Washing Appliances (WA).
Table 3: Customer preferences overview for different loads.
Load Customer preferences HP • Preferred temperatures
• DRCS
WA • Start time
• Task time
• DRCS
• CW
EV • Start time (when EV is parked)
• End time (When EV is leaving)
• DRCS
Table 4: Load run status (RS) description.
Run Status Description
0 The appliance must be turned OFF 1 The appliance must be turned ON 2 The appliance CAN operate
3.2 HEAT PUMP MATHEMATICAL MODELLING
To model the behavior of the HP a thermal household model is developed. In buildings the heat energy demand is calculated based on the energy flows, which can be divided into gains and losses. The gains include solar gains, internal heat gains, and active heat- ers. The losses include transmission and ventilation losses. For modelling purposes all the physical components of the building have to be considered and this might result in high-order differential equations according to [45]. In order to have a lower order model which is more suitable for the purposes of control algorithms grey or black box model- ling method is usually applied [46]. Modelling methods can be devided to white, grey and black box methods. In white box modelling the physical process is completely de- scribed. In black box methods the parameters have no physical meaning and are based on prior measurements, which is more suitable if no prior models exist [46]. Hence, grey box modelling method is more suitable here as they are based on physical proper- ties of the household. Similar approaches are used in [47, 48, 49].
An analogy between thermal and electrical systems (see Table 5) can be used to realize a thermal model of a household using a resistance capacitance (RC) circuit. In this mod- el a heat pump is used as the heat source to the household. The effects of solar gains by means of solar irradiance and radiation losses on the envelope are included in this mod- el, while internal heat gains are ignored. A typical household envelope is modelled with the following components (see Fig. 9):
Table 5: Thermal electrical analogy for thermal models of households.
Electrical Thermal
Parameter Unit Parameter Unit
Voltage V (V) Temperature (K)
Current I (A) Heat Flux Ф (W)
Stored charge q (C) Stored heat Q (J)
Electrical resistance Relec (Ω,V/A) Thermal resistance Rth (K/W) Electrical capacitance Celec (F,C/V) Thermal capacitance Cth (J/K)
Fig. 9: Simplified household thermal envelope.
The components of the household thermal model in (Fig. 9) are:
• 𝑅𝑡ℎ Thermal resistance that includes all thermal resistances in the household.
• 𝑅𝑜 Thermal resistance between outside envelope and air.
• 𝐶𝑣Thermal storage capacity.
• Ф𝐻𝐻Heat flux from HP.
• Ф𝑖𝑟𝑟 Heat flux from solar irradiance.
• Ф𝑒 Heat flux lost due to radiation.
• Ф𝑐 Heat flux inside the household.
• 𝑇𝑖𝑒 Temperature inside the house.
• 𝑇𝑜𝑜𝑡 Outside air temperature.
• 𝑇𝑡 Outside envelope surface temperature.
The discharging characteristics of a general RC circuit is shown in equations (1-2). Here it is assumed that the capacitor is first fully charged by the supply voltage before dis- charging its voltage to the resistance.
𝐶𝐶𝐶𝐶𝐶𝐶𝑡𝐶𝐶 𝐷𝐶𝐷𝐶ℎ𝐶𝐶𝑎𝐶𝑎𝑎 𝑉 =𝑉𝑡𝑒−𝑡𝜏 (1) 𝑇𝐶𝑇𝑒 𝐶𝐶𝑎𝐷𝑡𝐶𝑎𝑡 𝑓𝐶𝐶 𝑑𝐶𝐷𝐶ℎ𝐶𝐶𝑎𝐶𝑎𝑎 τ=𝑅𝐶 (2) Where:
V is the voltage across the capacitor.
To estimate the parameters in this model, a typical household with 180 m2 area is as- sumed with a heat pump of 5.5 kW thermal heat flux and a COP of 2.9 (1.89 kW elec- trical). The maximum temperature difference the HP can produce is assumed to be 45 K. Using temprature voltage analogy we get:
𝑅𝑡ℎ= 𝑇Ф𝑑𝑑𝑑𝑑
𝐻𝐻 =5500 𝑊45 𝐾 = 8.18 𝑇𝑊𝐾 Where:
𝑇𝑒𝑖𝑑𝑑 is the max temperature difference the HP produces (in K).
𝐶𝑣 can be estimated by assuming that the household losses all heat stored in 1.5 days (in J/K)
𝐶𝑣 =𝑅𝜏
𝑡ℎ= 24×1.5×3600
8.18×10−3 = 15.84 𝑀𝐾𝐽 (3)
To validate the estimations of 𝑅𝑡ℎ and 𝐶𝑣 with the mathematical calculations, the tem- perature change is plotted from the software model (Fig. 10). From boundary conditions of equations (1), the temperature value at 𝑡 =𝜏 should be 36.5% of the starting temper- ature for discharging. Here the outside temperature was fixed to -10 and the HP was not operated, and the inside temperature of the household was set to 35 oC. All the other gains and radiation losses were ignored here, just for the purposes of this check.
Fig. 10: Temperature discharge in a household.
𝑅𝑜 is estimated to be 4.0 𝑇𝑊𝐾 [50, p. 20]
Table 6: Household thermal model parameters.
Parameter Estimated value
𝑅𝑡ℎ 8.18 𝑇𝐾
𝐶
𝑅𝑜 4.0 𝑇𝐾
𝐶
𝐶𝑣 15.84 𝑀𝐽
𝐾 𝑃𝐻𝐻(𝑒𝑒𝑒𝐶𝑡𝐶𝐶𝐶𝐶𝑒) 1.98 𝑘𝐶
Next, the heat flux equations are solved as follow:
Ф𝑒 = 0.7𝜎𝐴𝑜(𝑇𝑡4− 𝑇𝑜𝑜𝑡4 ) (4)
Where:
𝜎 is Stefan-Boltzmann constant = 5.670367×10−8 W m−2 K−4. 𝐴𝑜 is the effective outside envelope area, assumed at 15 m2. 𝑇𝑜𝑜𝑡 is the outside air temperature (in K).
𝑇𝑡 is the outside envelope surface temperature (in K).
Ф𝑒 is the heat flux lost due to radiation. (in W)
Ф𝑖𝑟𝑟 = 0.7𝑃𝑖𝑟𝑟𝐴𝑤𝑖𝑒 (5)
Where:
𝑃𝑖𝑟𝑟 is the solar irradiance (in W m-2).
𝐴𝑤𝑖𝑒 is the windows area, assumed 10 m2.
Ф𝑜 =𝑇𝑠−𝑇𝑅𝑜𝑜𝑡
𝑜 (6)
Ф𝑡ℎ =Ф𝑜+Ф𝑒 − Ф𝑖𝑟𝑟 (7)
Ф𝑐 =Ф𝐻𝐻− Ф𝑡ℎ (8)
Where:
Ф𝐻𝐻is the heat flux from HP. (in W)
Ф𝑖𝑟𝑟 is the heat flux from solar irradiance. (in W) Ф𝑒 is the heat flux lost due to radiation. (in W) Ф𝑐 is the heat flux inside the household. (in W) 𝑇𝑖𝑒 is the temperature inside the house(in K).
𝑇𝑜𝑜𝑡 is the outside air temperature (in K).
𝑇𝑡 is the outside envelope surface temperature (in K).
After obtaining the heat flux inside the house (Ф𝑐), the new inside temperature will be:
𝑇𝑒𝑟𝑤 =𝑇𝑖𝑒+Ф𝐶𝑐
𝑣 ∆𝑡 (9)
Where:
𝑇𝑒𝑟𝑤 is the new inside temperature. (in K)
∆𝑡 is the time step. (in seconds)
Ф𝑐 is the heat flux inside the household. (in W) 𝑇𝑖𝑒 is the temperature inside the house. (in K)
The thermal house model hence can be viewed with a grey box method as in (Fig. 11), with 4 inputs and one output representing the new temperature inside the household.
The values of the outside temperature and solar irradiance is taken from historical re- cordings of from 2015. The data is obtained from the University of Applied Sciences in Wels PV and meteorology data recording system.
Fig. 11: Heat pump Grey box model.
3.3 HEAT PUMP CONSTRAINT
The HP has 2 operation status (𝑂𝑃𝐻𝐻𝑡 ) “on” or “off”. Once its operating (on state) it con- sumes the rated power 𝑃𝐻𝐻 kW. The temperature inside the house is the constraint and the HP operates to ensure that limits are kept.
𝑇𝑟𝑖𝑒 ≤ 𝑇𝑖𝑒≤ 𝑇𝑟𝑡𝑚 (10)
The algorithm flow chart is shown in (Fig. 12). First, the temperature limits are set. The limits depend on the customer preferences, which are the preferred temperature and DRCS. If the DRCS is set off then the inside temperature will be kept in a range of ±0.5 ℃ from the preferred temperature, and if it is set on the range will be±1 ℃ or as per customer’s request. The model output is the Run Status (𝑅𝑅𝐻𝐻𝑡 ) and its value will depend on the inside temperature. If the inside temperature is higher than the maximum limit, the HP must be turned OFF. If the inside temperature is lower than the maximum limit, the HP must be turned ON. And if the inside temperature is between the limits, the HP CAN operate.
𝑅𝑅𝐻𝐻𝑡 = 0 ( 𝑇𝑖𝑒 >𝑇𝑟𝑡𝑚) (11)
𝑅𝑅𝐻𝐻𝑡 = 1 ( 𝑇𝑖𝑒 <𝑇𝑟𝑖𝑒) (12)
𝑅𝑅𝐻𝐻𝑡 = 2 (𝑇𝑟𝑖𝑒 ≤ 𝑇𝑖𝑒≤ 𝑇𝑟𝑡𝑚) (13)
Fig. 12: Heat pump load constraint flowchart.
3.4 WASHING APPLIANCES MATHEMATICAL MODEL
The washing appliances (WA) models included in this thesis are the dishwasher (DW), tumble dryer (TD), and washing machine (WM) models. They are similar in structure and are only different in the values that their parameters take. To be specific they have different rated power consumptions, average task time, and operation times during the week days. The WAs are task based loads; they can be modeled by the customer setting a specific task.
To model the customer’s behavior for the WAs a statistical model is developed to speci- fy the starting time, end time, average power consumption, and frequency of operation during one week. According to statistical data from Austria [17], for (DW) the median power consumption for one year is 278 kWh, for (TD) is 162 kWh, and for (WM) is 236 kWh. Then according to average power consumption of each appliance the average task time and operation frequency during one week is chosen. This is modeled based on a normal distribution function with a mean (𝜇) and standard deviation (𝜎). For the commitment time, the day is divided into hour segments with constraints to insure a realistic customer behavior. That ensures that the washing is not finished in the middle of the night. Tasks between 4 and 7 PM will finish before midnight, between 7 and 9 PM will finish before midnight or in the next morning, and between 9 and 10 PM will finish in the next morning. The generated distribution histograms are shown in Appen- dix (Fig._A 1-Fig._A 9). The statistical model hence can be viewed as a black box method
Fig. 13: Washing appliances user behavior black box model.
Table 7: Operation Parameters for washing appliances.
Appliance type Power consumption (kW) Task time(H) Average operation time in one week
DW 𝜇= 1.2,𝜎= 0.1 𝜇 = 2,𝜎= 0.5 2-3
TD 𝜇= 1.8,𝜎= 0.3 𝜇 = 0.75,𝜎= 0.25 2-3
WM 𝜇= 0.5,𝜎= 0.1 𝜇 = 1.6,𝜎= 0.5 3-4
3.5 WASHING APPLIANCES CONSTRAINTS
The WA has a 2 operation status (𝑂𝑃𝑊𝑊𝑡 ) “on” or “off”. Once it is operating (on state) it consumes the rated power 𝑃𝑊𝑊 kW generated from the statistical model (Fig. 13). First the WA model structure will be explained and then the different parameters each one takes will be discussed. The WAs are task based appliances where the customers set up the starting time (𝑡𝑊𝑊𝑡𝑡𝑡𝑟𝑡), ending time (𝑡𝑊𝑊𝑟𝑒𝑒), and required time to operate (𝑡𝑊𝑊𝑡𝑡𝑡𝑡). The customer sets the value of the DRCS to indicate if they are willing to commit this appli- ance to DR mechanisms. If the DRCS is set to OFF then the WA will run immediately, and if it is set to ON the CEMS will decide when to run the WA within the specified commitment window (𝐶𝐶𝑊𝑊) which is calculated from 𝑡𝑊𝑊𝑡𝑡𝑡𝑟𝑡 and 𝑡𝑊𝑊𝑟𝑒𝑒 .Once the WA starts it has to finish the operation without any interruptions due to internal physical limitations concerning the quality of the washing or drying [51].
The algorithm flow chart is shown in (Fig. 14). The model output is the Run Status (𝑅𝑅𝑊𝑊𝑡 ) and its value will depend on time constraints. If the time is outside the 𝐶𝐶𝑊𝑊 or if the load already finished the task it must be OFF. If the load already started or if the remaining time in the 𝐶𝐶𝑊𝑊 is less than the task time the load must be ON. And if the WA haven’t started yet and there is enough time left in the 𝐶𝐶𝑊𝑊 the load CAN run.
𝑅𝑅𝑊𝑊𝑡 = 0 (𝑡𝑊𝑊𝑆𝑡𝑡𝑟𝑡 > 𝑡> 𝑡𝑊𝑊𝑟𝑒𝑒) or �𝑡𝑡=𝑡𝑊𝑊𝑒𝑒𝑑 ∆𝑡.𝑂𝑃𝑊𝑊𝑡
𝑊𝑊𝑆𝑡𝑆𝑆𝑡 =𝑡𝑊𝑊𝑡𝑡𝑡𝑡 (14)
𝑅𝑅𝑊𝑊𝑡 = 1 (𝑡𝑊𝑊𝑆𝑡𝑡𝑟𝑡 ≤ 𝑡 ≤ 𝑡𝑊𝑊𝑟𝑒𝑒) and 𝑂𝑃𝑊𝑊𝑡−1=1 or ( (𝑡𝑊𝑊𝑟𝑒𝑒− 𝑡) ≤ 𝑡𝑊𝑊𝑡𝑡𝑡𝑡) (15) 𝑅𝑅𝑊𝑊𝑡 = 2 (𝑡𝑊𝑊𝑆𝑡𝑡𝑟𝑡 ≤ 𝑡 ≤ 𝑡𝑊𝑊𝑟𝑒𝑒) and 𝑂𝑃𝑊𝑊𝑡−1=0 (16)
Fig. 14: Washing appliances Load constraint flowchart.
3.6 ELECTRIC VEHICLE MATHEMATICAL MODEL
There are three main factors considered when modelling an EV load profile charging location, charging moments, and charging need [52]. The charging location is the site where the EV is being charged (Home, Office, Parking lot). The charging moment rep- resents the plug in time of the EV and the parking time. The third factor is the charging need which represents the amount of electricity needed to charge the EV. This will de- pend on the initial state of charge (SoC) of the battery (Table 8).
Table 8: EV Modelling factors
Factor Modelling approach Charging location Household
Charging moments Based on arrival and leaving Time statistical distribution
Charging need • Initial SoC estimated based on driving distant statistical dis- tribution
• SoC change based on charging curve of a standard EV
3.6.1 CHARGING LOCATION
The EV is modelled so that it will only be charged in the household. The IEC defines different 4 modes of charging infrastructures [53]. In mode 1 the EV is directly con- nected to the AC mains, either via single or three phase connection at a maximum cur- rent of 16 A. Mode 2 is similar to mode 1 but with a maximum charging current of 32 A. In mode 3 the EV is connected to an electric vehicle supply equipment (EVSE) either via single or three phase connection including a control cable. Here currents up to 250 A can be used in the three phase configuration. Mode 4 is a direct fast charging DC mode with 600V and currents up to 400 A, the EV is also connected directly to an EVSE. The first option from (Table 9) is chosen as it’s the most common charging sta- tion used in typical households.
Table 9: Charging Infrastructure according to IEC 61851-1 [54].
Connection Mode Grid connection AC voltage AC current Type of charge
Mode 1 (AC) 1 phase 230 V 16 A Slow
3 phase 400 V 16 A Slow
Mode 2 (AC) 1 phase 230 V 32 A Slow
3 phase 400 V 32 A Slow
Mode 3 (AC) 1 phase 230 V 32 A Slow
3 phase 690 V 250 A Medium
Mode 4 (AC) - 600 V 400 A Fast
3.6.2 CHARGING MOMENTS
To model the customer’s behavior for the EV a statistical model is developed to specify the arrival time, leaving time, and driven trip distant. The arrival and leaving times de- pend on the day of the week and are modeled base on a normal distribution with a spe- cific mean (𝜇) and standard deviation (𝜎) (see Table 10). It is assumed that the customer has one trip per day. The generated distribution histograms are shown in Appendix (Fig._A 11-Fig._A 15). The difference between the arrival time and leaving time repre- sents the time where the car is parked in the house and hence the commitment window.
Table 10: Driving behavior parameters.
Day Leaving Time Arrival Time
Monday to Thursday 𝜇= 06: 30,𝜎= 00: 30 𝜇 = 17: 00,𝜎= 00: 30 Friday 𝜇= 06: 30,𝜎= 00: 30 𝜇 = 15: 00,𝜎= 00: 30 Saturday/Sunday 𝜇= 08: 30,𝜎= 00: 30 𝜇 = 11: 00,𝜎= 00: 30
Fig. 15: Electric vehicle user behavior black box model.
3.6.3 CHARGING NEED
Different approaches are commonly used to model the charging need such as using con- stant distance driven with constant electricity consumption level, SoC estimation with pre-defined distributions, trip distant and power consumption based on Gaussian distri- butions, or stochastic driving cycles to predict future behavior [52]. The problem can be split into two parts, the initial SoC estimation and the charging effect on the SoC once the EV is plugged.
To estimate the initial SoC, the distance driven for each trip is used based on statistical data (same model as in Fig. 15). According to statistical data from Austria [55] [56], the average driving length increased for men (from 15.4 km in 1995 to 17.5 km in 2008) and also for women (from 8.5 km in 1995 to 12.1 km in 2008). The average trip dura- tion is 23 minutes (see Fig. 16 for working day trips statistics). The length of workday trips is found to follow a Gamma distribution with a scale parameter (b) and a shape parameter (c). The scale parameter specifies the scale of the variate and the shape pa- rameter specifies shape of the variate. As almost 95% of the trips are less than 25 km, the parameters are chosen to be b=8 km and c=1.4 km. The generated distribution histo- gram is shown in Appendix (Fig._A 10).
𝑅𝐶𝐶% =�1−𝐷𝐷𝑡𝑆𝑑𝑡
𝑚𝑆𝑚�% (17)
Where 𝐷𝑡𝑟𝑖𝑡 is the total trip length in km, 𝐷𝑟𝑡𝑚 is the maximum distant the EV can drive.
